Face recognition result missing all predictions

I wrote an Image processing to MQTT service (will publish it soon) which allows users of it to define image datasources (different camera urls) and run against DeepStack container with scheduler, when it finds a face (or every 5 minutes) it send MQTT with the data so other systems will react to that.

MQTT message contains following information:

  • Status (Unknown / Identified / None)
  • Duration of processing
  • predictions
  • Image Url
  • Timestamp

I’m using Node Red for identifying who is at the main door and do some actions.

On top of it, it has more features I needed:

  1. Analytical data (number of images processed with duration per 1/5/15 minutes)
  2. Within the MQTT message link to the image processed (used for sending Telegram notification with the image)
  3. Person entity that represents one or more registered faces
  4. Manage faces - exposes DeepStack functionality to register / list / delete faces
  5. Test endpoint for faces based on image URL

I will release it once I’ll create managment UI (currently works as web API)

I encountered an issue when I have 2 registered faces for the same person and I’m sending an image of that person, I get only one predication with the best match, is it by design?

Hi @elad.bar Great work on this. DeepStack only returns the highest match of a face. Are you asking for it to return like top 5 matches?

Hi @john, that would be great if feasible, maybe with cofiguration so it will not affect other current integrations available

1 Like

That’s a good idea. It is entirely feasible. A possible design of it is to add an extra parameter to each request that specifies how many results to return per face.